Research map / Vision-language and multimodal / CLIP and zero-shot
Intra-Modal Proxy Learning for Zero-Shot Visual Categorization with CLIP
Qi Qian, Yuanhong Xu, Juhua Hu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| learn intra-modal vision proxy (InMaP) instead of text proxy | CLIP zero-shot · ImageNet ResNet-50 | accuracy (%): 60.32 -> 63.74 | Improved |
| InMaP on large encoder | CLIP ViT-L/14@336 · ImageNet | accuracy (%): 77.02 -> 80.21 | Improved |
| add Sinkhorn-refined pseudo labels (ablation) | InMaP^0.5 · ImageNet ResNet-50 | accuracy (%): 60.95 -> 63.74 | Improved |
| Sinkhorn labels without proxy learning (ablation) | CLIP baseline · ImageNet ResNet-50 | accuracy (%): 60.32 -> 62.53 | Improved |
| proxy learning without refined labels (ablation) | CLIP baseline · ImageNet ResNet-50 | accuracy (%): 60.32 -> 60.83 | Improved |
| raise tau_I from 0.01 to 0.04 (ablation) | InMaP · ImageNet ResNet-50 | accuracy (%): 62.91 -> 63.74 | Improved |
| raise tau_I from 0.04 to 0.05 (ablation) | InMaP · ImageNet ResNet-50 | accuracy (%): 63.74 -> 63.29 | Hurt |
| make all pseudo labels one-hot (alpha=0) (ablation) | InMaP · ImageNet ResNet-50 | accuracy (%): 63.74 -> 63.14 | Hurt |
| vary threshold alpha between 1 and 0.5 (ablation) | InMaP · ImageNet ResNet-50 | accuracy (%): 63.49 -> 63.72 | No clear effect |
| gamma=1 (no label distribution balancing) versus gamma=0.1 (ablation) | InMaP · ImageNet ResNet-50 | accuracy (%): 63.75 -> 60.95 | Hurt |
| single prompt versus ensemble (ablation) | InMaP · ImageNet ResNet-50 | accuracy (%): 63.15 -> 63.74 | Mixed |
| single prompt versus ensemble (ablation) | Baseline · ImageNet ResNet-50 | accuracy (%): 58.15 -> 60.32 | Improved |
Closest papers by meaning
- Interpreting and Analysing CLIP's Zero-Shot Image Classification via Mutual Knowledge (NeurIPS 2024 · similarity 0.89)
- Understanding Transferable Representation Learning and Zero-shot Transfer in CLIP (ICLR 2024 · similarity 0.89)
- Language-Driven Cross-Modal Classifier for Zero-Shot Multi-Label Image Recognition (ICML 2024 · similarity 0.88)
- TPR: Topology-Preserving Reservoirs for Generalized Zero-Shot Learning (NeurIPS 2024 · similarity 0.87)
- Building Vision-Language Models on Solid Foundations with Masked Distillation (CVPR 2024 · similarity 0.87)
- Mining Fine-Grained Image-Text Alignment for Zero-Shot Captioning via Text-Only Training (AAAI 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
